Product
Vladimir Bazalenko, Founder, RendixUpdated Oct 9, 2026AI prompt research finds the questions buyers ask AI assistants about your category, and the searches the engines run to answer them.
Rendix asks 150 buyer questions about your category on a live engine, sorts them into topics, shows which topics no brand owns yet, and records the searches each engine ran for the questions you track.
7-day free trial. Cancel anytime.
Profound came first in 20 % of the answers, so no brand owns this topic. Sites cited most: blog.hubspot.com, ziptie.dev, reddit.com, seranking.com.
Definition
It samples the questions and reads the answers. What that can and cannot tell you:
No AI platform publishes how often a question is asked, so every prompt volume is modelled.
× 0.115 is what Ahrefs multiplies a keyword's Google volume by to estimate its ChatGPT volume.
Ahrefs says it does not claim a validated link between Google volume and how often a question is asked in an AI tool. Scrunch notes that no platform offers a first-party prompt feed, and panels miss mobile apps and Safari.
Source: Ahrefs, Sep 2026 · Scrunch, Sep 2026
Asked once each, buyer questions show who the engines name first in each topic.
6 of 6 topics in Rendix's own category had no owner: no brand came first in more than 20 % of the answers.
150 generated buyer questions, 25 a topic, asked once each on Perplexity with live web, Sep 4, 2026. A topic counts as owned when one brand comes first in 40 % of its answers.
Source: Rendix market sample, Sep 2026
Buyers ask in sentences. The engines search in keywords.
1.2 % of ChatGPT's searches were phrased as a question; Perplexity 0.7 %, Gemini 2.2 %.
2,357 searches behind the answers to 36 questions, Aug 10 to Oct 7, 2026. A full bar is 10 %. AI Overviews shows no searches.
Source: Rendix scans of its own category
Method
A sample of the market, taken on a live engine and sorted into topics, every number with its count.
A model writes 150 buyer questions from your brand profile, with no brand names in them, of kinds such as best for, compare, recommend and alternatives.
Every question goes to Perplexity with live web search. Rendix keeps the brands the answer names, in order, and the pages it cites.
Questions are grouped into topics. For each one: the brands named most, the brand that comes first most often, and the sites cited.
Fewer than 10 answers: not enough to say. Your brand named: present. One brand first in 40 % or more: contested. Otherwise: open.
Move the questions worth watching into tracking, where all four engines answer them three times a week. A deep scan asks 30 more questions on one topic.
| Topic, 25 questions each | Named most | Leader first in | Verdict | Cited most |
|---|---|---|---|---|
| AI visibility tracking | Profound, 18 | 20 % | Open | blog.hubspot.com |
| Citation monitoring | Profound, 13 | 16 % | Open | airops.com |
| AI reputation management | Profound, 8 | 8 % | Open | llmpulse.ai |
| Generative engine optimization | Profound, 8 | 16 % | Open | linkedin.com |
| Brand intelligence analytics | Profound, 10 | 20 % | Open | tryanalyze.ai |
| Answer engine audits | Scrunch, 7 | 8 % | Open | linkedin.com |
Fan-out
Query fan-out is the set of web searches an AI engine runs to answer one question. Google defines it as "a set of concurrent, related queries generated by the model", and ChatGPT, Perplexity and Gemini do the same without using the name.
Rendix stores the searches each engine actually ran, when the engine shows them. Most fan-out tools write likely searches with a model instead, so a captured search and a predicted one are different evidence.
The searches change from run to run: only 5.3 % of Gemini's repeated word for word on the next run. Google warns that separate content for every fan-out variation, made to steer its AI answers, breaks its spam policy. Rendix uses the searches to check whether one of your pages already answers them.
Source: Google Search Central, Jul 2026
leading solutions for measuring brand share of voice in ai
| Engine | Searches shown | Per answer | As a question | With a brand | With a year | Same next run |
|---|---|---|---|---|---|---|
| ChatGPT | 61 of 748 | 3 | 1.2 % | 54.2 % | 28.6 % | 0 of 89 |
| Perplexity | All from Sep 28 | 3 | 0.7 % | 4.4 % | 22.5 % | 3.0 % |
| Gemini | 630 of 777 | 2 | 2.2 % | 1.9 % | 1.9 % | 5.3 % |
| AI Overviews | Never | - | - | - | - | - |
In Rendix
The sample, the topics it sorts into, and the searches behind the questions you track.
150 fresh buyer questions, 2 runs a month on paid plans, with the brands named and the sites cited for each.
Research · Market panelEach topic's verdict next to what your tracked questions on it measure, kept apart so a sample never mixes with tracking.
Research · topics tableThe searches each engine ran for every question you track, grouped under the question, with whether one of your pages answers them.
Prompts → Query fanouts30 more questions on one topic, 5 a month on paid plans, when a topic needs a closer look.
A topic's drawerThe brands named instead of you on each question, and their share of voice.
CompetitorsThe pages to write or fix for the questions you lose, each with a brief.
ActionsPlans
Straight from the pricing table: only what ships today.
Questions
/ Updated Oct 9, 2026Need more help? team@rendix.io
Finding the questions buyers ask AI assistants about your category, and what the engines answer and search for each one.
Rendix asks 150 generated buyer questions on a live engine, groups them into topics, and shows which topics no brand owns yet. For the questions you track, it records the searches each engine ran.
No one can. AI platforms publish no query counts.
Prompt volumes come from browser panels or from Google search volume multiplied by a ratio: Ahrefs multiplies by 0.115 for ChatGPT and says the link is not validated. Rendix shows no volume number; it ranks questions by what the answers show: who is named, and where you are missing.
Source: Ahrefs, Sep 2026 · Scrunch, Sep 2026
The set of web searches an AI engine runs to answer one question.
Google uses the term for AI Mode and AI Overviews; ChatGPT, Perplexity and Gemini do the same. In Rendix's own category a typical answer ran two to three searches, written as keywords: 0.7 to 2.2 % were phrased as questions.
Source: Google Search Central, Jul 2026
Only when ChatGPT shows them, which is not every answer.
In Rendix's scans ChatGPT exposed its searches in 61 of 748 answers. Perplexity and Gemini return their searches with the answer; Google AI Overviews shows none. Rendix stores the searches an engine actually ran and never fills the gap with generated ones.
No. Google calls that scaled content abuse.
Its guide says creating separate content for every variation, fan-out queries included, "primarily to manipulate rankings or generative AI responses" violates its spam policy. The searches also change between runs: only 5.3 % of Gemini's repeated word for word.
Source: Google Search Central, Jul 2026
No. They are a sample written by a model.
Every prompt research tool writes its questions with a model or estimates them from panels; none sees real prompts. Rendix calls each run a sample, asks every question on a live engine, and keeps the date and the engine with each answer.
When your market or your product changes, and at least monthly.
Rendix allows 2 market samples and 5 deep scans a month on paid plans. The questions you move into tracking are asked on all four engines three times a week.
Get started
/ Free 7-day trialOne scan shows your place in AI answers – and the first moves to climb. Free, no account needed.